How to Write Better AI Prompts: Our 6-Part Framework

Difficulty: Beginner | Estimated time: 15 minutes

---

The difference between someone who gets mediocre results from AI and someone who gets genuinely useful output almost always comes down to prompting. Not the tool. Not the model. The prompt.

We have written thousands of prompts across dozens of AI tools over the past two years. Along the way, we developed a 6-part framework that consistently produces better output. It works with any AI tool: chatbots, coding assistants, image generators, writing tools, you name it.

Here is the framework, step by step, with real examples.

Step 1: Set the Role

Start your prompt by telling the AI who it is. This sounds strange, but it works. When you assign a role, the AI adjusts its tone, vocabulary, depth, and assumptions to match.

Weak prompt: "Write me a blog post about email marketing."

Strong prompt: "You are a senior email marketing strategist with 10 years of experience working with SaaS companies. Write a blog post about email marketing."

The second prompt produces output that is more specific, more confident, and uses industry language naturally. The role acts as a filter on the AI's enormous knowledge base, narrowing it to what is relevant.

For project planning and brainstorming prompts, pairing a well-set role with a visual tool like MindManager helps you map out the AI's output into actionable workflows.

> What to look for: Match the role to the expertise you need. "You are a helpful assistant" is the default and gives you default output. Be specific.

Step 2: Give Context

Context is the background information the AI needs to produce relevant output. Without it, the AI guesses, and guesses are generic.

Context includes: who the audience is, what has already been tried, what constraints exist, what the goal is, and any relevant facts the AI would not know on its own.

Example: "Our audience is technical founders at pre-seed startups. They are skeptical of marketing advice and respond best to data-driven arguments. The blog post will be published on our company blog, which gets around 5,000 monthly readers."

The more context you provide, the less the AI has to assume. Assumptions are where output quality breaks down. You do not need to write a novel, but three to five sentences of context makes a measurable difference.

Step 3: State the Task

Now tell the AI exactly what you want it to do. This should be a clear, single instruction. If you need multiple things, break them into separate prompts or number them explicitly.

Weak: "Help me with our newsletter."

Strong: "Write a 400-word newsletter introduction that announces our new API feature, explains why it matters to developers, and includes a call-to-action linking to our documentation."

Notice the specifics: word count, content requirements, audience, and desired action. Every detail you include is a detail the AI does not have to guess.

Step 4: Specify the Format

Tell the AI what the output should look like. This is especially important for anything you plan to use directly without heavy editing.

Format specifications include: length (word count, number of bullet points), structure (headers, numbered list, table), tone (formal, conversational, technical), and any formatting requirements (markdown, HTML, plain text).

Example: "Format the output as a markdown document with H2 headers for each section. Use bullet points for lists. Keep paragraphs to 3 sentences maximum. Total length: 600-800 words."

Tools like PopAI can help you structure and present AI-generated content visually, which is useful when the output needs to go into presentations or documents.

> What to look for: If the AI ignores your format instructions, repeat them at the end of the prompt. Some models weight the end of a prompt more heavily than the middle.

Step 5: Add Examples

Examples are the most powerful and most underused part of prompting. When you show the AI what good output looks like, it patterns-matches against your example rather than its general training data.

You can provide examples in two ways:

Input-output pairs: "Here is an example of the kind of email subject line I want: Input: New feature launch for project management tool. Output: Ship projects 2x faster: our new timeline view is live."

Style references: "Here is a paragraph from our blog that represents the tone we want: [paste paragraph]. Match this tone in your output."

Even one example dramatically improves consistency. Three examples make the output remarkably reliable.

For building reusable prompt templates with examples, we use Crush to create and manage content workflows that include embedded prompt frameworks.

Step 6: Iterate

Your first prompt will rarely produce perfect output. That is expected and fine. The key is to iterate intentionally rather than starting from scratch.

When the output is close but not right, do not rewrite the entire prompt. Instead, add a follow-up that targets the specific problem: "Good, but make the tone more conversational," "Shorten each section by half," "Replace the generic examples with ones specific to e-commerce."

Keep a record of what worked. After a few iterations, you will have a refined prompt that produces consistent results. Save that prompt. It is now a reusable asset.

Read more about prompting strategies in our guides at How to Prompt AI Agents and browse the full Agent Tutorials category.

Putting It All Together

Here is the full framework in one prompt:

"You are a [role]. [Context about the situation, audience, and goals]. [Specific task with clear deliverable]. [Format: length, structure, tone]. Here is an example of what good output looks like: [example]. [Any additional constraints]."

That is it. Six parts. It takes 60 seconds longer to write than a vague prompt and produces output that is dramatically more useful. Start using this framework today, and within a week you will wonder how you ever prompted without it.